{
  "model_type": "nanochat_gpt",
  "architecture": "GPT with Rotary Embeddings, QK Norm, and ReLU^2 activation",
  "vocab_size": 65536,
  "sequence_len": 512,
  "n_layer": 20,
  "n_head": 10,
  "n_kv_head": 10,
  "n_embd": 1280,
  "torch_dtype": "float32"
}